On 8 November 2010 17:32, Reece Dunn <[email protected]> wrote: > On 8 November 2010 04:45, Austin Lund <[email protected]> wrote: >> On 8 November 2010 11:49, James McKenzie <[email protected]> wrote: >>> Thus a second test case needs to be >>> developed that is only for Windows7 and the remaining test skipped for >>> Windows7. Something like what we do for Unicode tests for Windows9x/ME. >> >> Isn't the rule that the tests should only check windows versions >> before testing behaviour if that is what applications to a similar >> thing, otherwise the test isn't required? > > The rule is not to check Windows version, but to infer the DLL version > (e.g. by checking for methods only available in newer versions of > Windows). That way, installing a new version of shlwapi.dll via > Internet Explorer on older systems does not break those tests.
That seems like a test which is necessarily true, but not sufficient. If the DLL ABI does not change the tests results might. In the spirit of the rule described on the wiki (and quoted above), it would seem highly unusual for a program to test for the presence or absence of an unrelated function before depending on some result. The use of broken() has taken me a while to understand. I believe it is better understood as not "broken" in particular, but a behaviour that wine should not reproduce but sometimes windows might (or does).
